A systematic review, meta-analysis, and network analysis of diagnostic microRNAs in glaucoma

PubMed, Web of Science, EMBASE, OVID, and Scopus databases were utilized, and papers from inception to December 30 2021 with the combination of the following search terms, were extracted: human, case or patient, control, microRNA, glaucoma, intraocular pressure, or hypertension. The full search strategy was shown in the method of literature search part.

("human" OR "humans" OR "patient" OR "patients" OR "control" OR "controls" OR "group" OR "groups") AND ("microrna" OR "micro RNA" OR "micrornas" OR "mirs" OR "microRNA") AND ("glaucoma" OR "intraocular hypertension" OR "intraocular pressure" OR "ocular hypertension").

Articles are eligible for inclusion if they fulfill the following criteria: [1] original articles that are designed as a case–control study to assess the microRNA expression profiling in glaucoma patients; [2] the study must provide a source of samples obtained from the case and control groups [3]. MicroRNA profiling must be accomplished using one of the qPCR, microarray, or next-generation sequencing methods, and [4] the cut-off criteria for detection of differentially expressed microRNA must be reported. Review articles, studies not published in English, and animal and cell line studies were excluded.

Two independent researchers (HS and MSS) screened, evaluated, and extracted the data to minimize the risk of selection bias in the study. In cases where two researchers disagreed on the inclusion or exclusion of an article, the third researcher (MR) made the definitive decision.

The quality of the included studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria [10]. This method uses seven questions to evaluate the diagnostic accuracy studies in four key domains: patient selection, index test, reference standard, flow, and time. The seven questions are listed as follows: (1) Could the selection of patients have introduced bias? (2) Are there concerns that the included patients and setting do not match the review question? (3) Could the conduct or interpretation of the index test have introduced bias? (4) Are there concerns that the index test, its conduct, or its interpretation differ from the review question? (5) Could the reference standard, its conduct, or its interpretation have introduced bias? (6) Are there concerns that the target condition, as defined by the reference standard, does not match the question? (7) Could the patient flow have introduced bias? The first two address the assessment of patient selection; the next two address the index test; questions five and six address the reference standard; and the last one addresses the flow and timing.

The primary search and study selection procedure is briefly demonstrated as a diagram, famously known as Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), in Fig. 1 [14]. As shown, our search strategy retrieved a total of 648 studies. After removing 327 duplicates, 321 studies remained for the title/abstract screening. Respectively, 284 irrelevant studies were omitted, and out of the remaining 37 studies, 6 met the inclusion criteria and entered our study [15‐20]. These case–control studies measured the microRNA levels in glaucoma patients and healthy people; two of them analyzed the aqueous humor [16, 20], two assessed tears [18, 19], and the other two investigated serum and plasma samples [15, 17]. Table 1 displays the extracted data from the included studies. While one study included only female participants [15], the rest assessed both females and males. Moreover, real-time polymerase chain reaction (RT-PCR), next-generation sequencing, and microarray systems were administered. The risk of bias analysis of the articles based on the QUADAS-2 checklist is summarized in Figs. 2 and 3.

After the screening of the 321 articles, finally, 6 of them were included [15‐20]. Fifty-two differentially expressed mature microRNAs in these selected studies compared 115 glaucoma patients with 108 controls. The total number of individual microRNAs found in the included studies was 52, of which 24 were down-regulated, and 28 were up-regulated. The diagnostic accuracy of miR-210-3 was measured in two different scenarios: a screening set with a smaller sample size, followed by a validation in a larger group of patients; it was up-regulated in both steps [17]. Of the 52 individual microRNAs, 3, 3, 11, and 36 were taken from serum, plasma, tears, and aqueous humor, respectively.

All the patients' plasma and serum microRNA samples were up-regulated [15, 17]. However, most of the patients' tear samples contained microRNAs that were up-regulated [18, 19] compared with the control group (9 versus 2). In contrast, of the 36 samples of aqueous humor microRNAs, 22 were down-regulated and 14 were up-regulated [16, 20].

The meta-analysis included studies that detected a single dysregulated microRNA in glaucoma cases compared to controls and the corresponding AUC, sensitivity, and specificity for the discovered microRNA. Using these criteria, four studies identified 12 individual microRNAs. The associated forest plot depicts the sensitivity (95% CIs) and specificity (95% CIs) for each microRNA (Fig. 4). Overall sensitivity and specificity of the 12 individual microRNAs in the diagnosis of glaucoma were 80% [95% CI 73–86%, test of heterogeneity: Q = 15.29, I² = 28.08%, p-value = 0.17] and 74% [95% CI 63–83%, test of heterogeneity: Q = 28.15, I² = 60.9%, p-value = 0.00], respectively. These findings suggest that utilizing microRNAs as biomarkers has a high discriminative ability in detecting glaucoma. For glaucoma detection, the slope coefficient was related to a p-value of 0.83 (Fig. 5), indicating no publication bias in our meta-analysis. In addition, the pooled diagnostic odds ratio (DOR) of the 12 individual microRNAs in the diagnosis of glaucoma was 11.2 [95% CI 5.3–24]. Afterward, random-effects models were used to re-analyze the data, and the diagnostic threshold was investigated. The Spearman's correlation coefficient was − 0.130 (p-value = 0.687), indicating that the diagnostic threshold was not responsible for the heterogeneity. Disparities in study techniques, specimen type, endogenous reference, or total sample size may also contribute to the existing heterogeneity. The publication bias of meta-analysis in diagnostic accuracy was investigated using Deeks' funnel plot asymmetry test, depicted in Fig. 6. The most well-known reporting bias is publication bias. It is the outcome of relevant trials being published or not, depending on the type and direction of the results. A paper, for instance, is more likely to be published if the findings are significant. Following subgroup analysis, it has been apparent that the most sensitive microRNA was mir-486-5p from aqueous humor (100%, 95% CI 66–100%). The most specific microRNAs were sampled from the plasma (mir-637, mir-1306-5p, and mir-3159). The data for subgroup analysis are provided in Additional file 1.